Examples with NodeManager org.apache.hadoop.hdds.scm.node.NodeManager used on opensource projects

Example 1 with NodeManager

use of org.apache.hadoop.hdds.scm.node.NodeManager in project ozone by apache.

the class TestReconAsPassiveScm method testDatanodeRegistrationAndReports.

@Test
public void testDatanodeRegistrationAndReports() throws Exception {
    ReconStorageContainerManagerFacade reconScm = (ReconStorageContainerManagerFacade) cluster.getReconServer().getReconStorageContainerManager();
    StorageContainerManager scm = cluster.getStorageContainerManager();
    PipelineManager reconPipelineManager = reconScm.getPipelineManager();
    PipelineManager scmPipelineManager = scm.getPipelineManager();
    LambdaTestUtils.await(60000, 5000, () -> (reconPipelineManager.getPipelines().size() >= 4));
    // Verify if Recon has all the pipelines from SCM.
    scmPipelineManager.getPipelines().forEach(p -> {
        try {
            assertNotNull(reconPipelineManager.getPipeline(p.getId()));
        } catch (PipelineNotFoundException e) {
            Assert.fail();
        }
    });
    // Verify we can never create a pipeline in Recon.
    LambdaTestUtils.intercept(UnsupportedOperationException.class, "Trying to create pipeline in Recon, which is prohibited!", () -> reconPipelineManager.createPipeline(RatisReplicationConfig.getInstance(ONE)));
    ContainerManager scmContainerManager = scm.getContainerManager();
    assertTrue(scmContainerManager.getContainers().isEmpty());
    // Verify if all the 3 nodes are registered with Recon.
    NodeManager reconNodeManager = reconScm.getScmNodeManager();
    NodeManager scmNodeManager = scm.getScmNodeManager();
    assertEquals(scmNodeManager.getAllNodes().size(), reconNodeManager.getAllNodes().size());
    // Create container
    ContainerManager reconContainerManager = reconScm.getContainerManager();
    ContainerInfo containerInfo = scmContainerManager.allocateContainer(RatisReplicationConfig.getInstance(ONE), "test");
    long containerID = containerInfo.getContainerID();
    Pipeline pipeline = scmPipelineManager.getPipeline(containerInfo.getPipelineID());
    XceiverClientGrpc client = new XceiverClientGrpc(pipeline, conf);
    runTestOzoneContainerViaDataNode(containerID, client);
    // Verify Recon picked up the new container that was created.
    assertEquals(scmContainerManager.getContainerIDs(), reconContainerManager.getContainerIDs());
    GenericTestUtils.LogCapturer logCapturer = GenericTestUtils.LogCapturer.captureLogs(ReconNodeManager.LOG);
    GenericTestUtils.setLogLevel(ReconNodeManager.LOG, Level.DEBUG);
    reconScm.getEventQueue().fireEvent(CLOSE_CONTAINER, containerInfo.containerID());
    GenericTestUtils.waitFor(() -> logCapturer.getOutput().contains("Ignoring unsupported command closeContainerCommand"), 1000, 20000);
}

Example 2 with NodeManager

use of org.apache.hadoop.hdds.scm.node.NodeManager in project ozone by apache.

the class TestReconAsPassiveScm method testReconRestart.

@Test
public void testReconRestart() throws Exception {
    final OzoneStorageContainerManager reconScm = cluster.getReconServer().getReconStorageContainerManager();
    StorageContainerManager scm = cluster.getStorageContainerManager();
    // Stop Recon
    ContainerManager scmContainerManager = scm.getContainerManager();
    assertTrue(scmContainerManager.getContainers().isEmpty());
    ContainerManager reconContainerManager = reconScm.getContainerManager();
    assertTrue(reconContainerManager.getContainers().isEmpty());
    LambdaTestUtils.await(60000, 5000, () -> (reconScm.getScmNodeManager().getAllNodes().size() == 3));
    cluster.stopRecon();
    // Create container in SCM.
    ContainerInfo containerInfo = scmContainerManager.allocateContainer(RatisReplicationConfig.getInstance(ONE), "test");
    long containerID = containerInfo.getContainerID();
    PipelineManager scmPipelineManager = scm.getPipelineManager();
    Pipeline pipeline = scmPipelineManager.getPipeline(containerInfo.getPipelineID());
    XceiverClientGrpc client = new XceiverClientGrpc(pipeline, conf);
    runTestOzoneContainerViaDataNode(containerID, client);
    assertFalse(scmContainerManager.getContainers().isEmpty());
    // Close a pipeline
    Optional<Pipeline> pipelineToClose = scmPipelineManager.getPipelines(RatisReplicationConfig.getInstance(ONE)).stream().filter(p -> !p.getId().equals(containerInfo.getPipelineID())).findFirst();
    assertTrue(pipelineToClose.isPresent());
    scmPipelineManager.closePipeline(pipelineToClose.get(), false);
    // Start Recon
    cluster.startRecon();
    // Verify if Recon has all the nodes on restart (even if heartbeats are
    // not yet received).
    NodeManager reconNodeManager = reconScm.getScmNodeManager();
    NodeManager scmNodeManager = scm.getScmNodeManager();
    assertEquals(scmNodeManager.getAllNodes().size(), reconNodeManager.getAllNodes().size());
    // Verify Recon picks up new container, close pipeline SCM actions.
    OzoneStorageContainerManager newReconScm = cluster.getReconServer().getReconStorageContainerManager();
    PipelineManager reconPipelineManager = newReconScm.getPipelineManager();
    assertFalse(reconPipelineManager.containsPipeline(pipelineToClose.get().getId()));
    LambdaTestUtils.await(90000, 5000, () -> (newReconScm.getContainerManager().containerExist(ContainerID.valueOf(containerID))));
}

Example 3 with NodeManager

use of org.apache.hadoop.hdds.scm.node.NodeManager in project ozone by apache.

the class TestQueryNode method testNodeOperationalStates.

@Test
public void testNodeOperationalStates() throws Exception {
    StorageContainerManager scm = cluster.getStorageContainerManager();
    NodeManager nm = scm.getScmNodeManager();
    // Set one node to be something other than IN_SERVICE
    DatanodeDetails node = nm.getAllNodes().get(0);
    nm.setNodeOperationalState(node, DECOMMISSIONING);
    // All nodes should be returned as they are all in service
    int nodeCount = scmClient.queryNode(IN_SERVICE, HEALTHY, HddsProtos.QueryScope.CLUSTER, "").size();
    assertEquals(numOfDatanodes - 1, nodeCount);
    // null acts as wildcard for opState
    nodeCount = scmClient.queryNode(null, HEALTHY, HddsProtos.QueryScope.CLUSTER, "").size();
    assertEquals(numOfDatanodes, nodeCount);
    // null acts as wildcard for nodeState
    nodeCount = scmClient.queryNode(IN_SERVICE, null, HddsProtos.QueryScope.CLUSTER, "").size();
    assertEquals(numOfDatanodes - 1, nodeCount);
    // Both null - should return all nodes
    nodeCount = scmClient.queryNode(null, null, HddsProtos.QueryScope.CLUSTER, "").size();
    assertEquals(numOfDatanodes, nodeCount);
    // No node should be returned
    nodeCount = scmClient.queryNode(IN_MAINTENANCE, HEALTHY, HddsProtos.QueryScope.CLUSTER, "").size();
    assertEquals(0, nodeCount);
    // Test all operational states by looping over them all and setting the
    // state manually.
    node = nm.getAllNodes().get(0);
    for (HddsProtos.NodeOperationalState s : HddsProtos.NodeOperationalState.values()) {
        nm.setNodeOperationalState(node, s);
        nodeCount = scmClient.queryNode(s, HEALTHY, HddsProtos.QueryScope.CLUSTER, "").size();
        if (s == IN_SERVICE) {
            assertEquals(5, nodeCount);
        } else {
            assertEquals(1, nodeCount);
        }
    }
}

Example 4 with NodeManager

use of org.apache.hadoop.hdds.scm.node.NodeManager in project ozone by apache.

the class TestReconIncrementalContainerReportHandler method testProcessICRStateMismatch.

@Test
public void testProcessICRStateMismatch() throws IOException {
    // Recon container state is "OPEN".
    // Replica state could be any Non OPEN state.
    long containerId = 11;
    for (State state : Arrays.asList(State.CLOSING, State.QUASI_CLOSED, State.CLOSED)) {
        ContainerWithPipeline containerWithPipeline = getTestContainer(containerId++, OPEN);
        ContainerID containerID = containerWithPipeline.getContainerInfo().containerID();
        ReconContainerManager containerManager = getContainerManager();
        containerManager.addNewContainer(containerWithPipeline);
        DatanodeDetails datanodeDetails = containerWithPipeline.getPipeline().getFirstNode();
        NodeManager nodeManagerMock = mock(NodeManager.class);
        when(nodeManagerMock.getNodeByUuid(any())).thenReturn(datanodeDetails);
        IncrementalContainerReportFromDatanode reportMock = mock(IncrementalContainerReportFromDatanode.class);
        when(reportMock.getDatanodeDetails()).thenReturn(containerWithPipeline.getPipeline().getFirstNode());
        IncrementalContainerReportProto containerReport = getIncrementalContainerReportProto(containerID, state, datanodeDetails.getUuidString());
        when(reportMock.getReport()).thenReturn(containerReport);
        ReconIncrementalContainerReportHandler reconIcr = new ReconIncrementalContainerReportHandler(nodeManagerMock, containerManager, SCMContext.emptyContext());
        reconIcr.onMessage(reportMock, mock(EventPublisher.class));
        assertTrue(containerManager.containerExist(containerID));
        assertEquals(1, containerManager.getContainerReplicas(containerID).size());
        LifeCycleState expectedState = getContainerStateFromReplicaState(state);
        LifeCycleState actualState = containerManager.getContainer(containerID).getState();
        assertEquals(String.format("Expecting %s in " + "container state for replica state %s", expectedState, state), expectedState, actualState);
    }
}

Example 5 with NodeManager

use of org.apache.hadoop.hdds.scm.node.NodeManager in project ozone by apache.

the class TestKeyInputStream method getNodeHealth.

private HddsProtos.NodeState getNodeHealth(DatanodeDetails dn) {
    HddsProtos.NodeState health = null;
    try {
        NodeManager nodeManager = getCluster().getStorageContainerManager().getScmNodeManager();
        health = nodeManager.getNodeStatus(dn).getHealth();
    } catch (NodeNotFoundException e) {
        fail("Unexpected NodeNotFound exception");
    }
    return health;
}